MRMS-based Hail Sizing and Classification Using Different, Large Databases

Storm Data is the primary record of severe weather events in the United States, like hail. However, many previous evaluations of Storm Data hail reports reveal that the reports have limited applicability towards developing hail sizing algorithms due to large uncertainties and inaccuracies in report location and magnitude, and the sparsity of the reports. A major effort ongoing at CIMMS/NSSL has been using the Severe Hazards Analysis and Verification Experiment (SHAVE) to improve hail sizing and classification algorithms for both polarimetric radars and the Multi-Radar, Multi-Sensor (MRMS) system. The SHAVE data set includes 731 cases from the ten years of operations and the reports are at high-resolution, comparable to the MRMS horizontal grid spacing. For this presentation, vertical reflectivity profiles from MRMS combined with near-storm environment (NSE) data have been paired to both Storm Data and SHAVE reports. The Storm Data reports span the years 2005 through 2011 and the MRMS data comes from the Multi-Year Reanalysis of Remotely Sensed Storms (MYRORSS). The vertical profiles and associated NSE data have been processed through different machine learning techniques to evaluate the ability of both data sets and techniques to produce an algorithm that can accurately size or classify hail. The differences in algorithm skill using Storm Data as the training set, compared to training with SHAVE, will be discussed with respect to large databases and data quality.